Lip feature analyzing in speech synthesis system for speech impaired

Abstract

A speech synthesis system based on lipreading technique was investigated for rehabilitation. In order to acquire more features of lip contours and improve the recognizing rate of lipreading, a new unsymmetrical lip contour model was analyzed to extract the degree of pouting, and the differential coefficients of some parameters were calculated to describe dynamic character of the lip contour. Experiments were carried out in a small database of isolated Chinese phonemes based on Hidden Markov Model (HMM) recognizer. Compared to the traditional symmetrical lip contour model, the new parameters from the unsymmetrical lip contour model improve the recognizing rate more than 25%. © 2011 Springer-Verlag Berlin Heidelberg.